 | | | AA and motion blurring a depth pass or not (was RE: DOF and Depth Pass probl | AA and motion blurring a depth pass or not (was RE: DOF and Depth Pass probl 2005-03-22 - By Brad Friedman
Back
you don't want to motion blur the depth pass as you'll get false depth
values. But you do run into the problem of which to render first... DOF
or Motion Blur. And I suppose this is why MR has the capability to
render both physically accurate (at serious performance cost but well...
you gotta do what you gotta do). There's no good answer to that one.
If you do a non AA'd DOF on a non AA'd beauty pass you get proper DOF
but now that the pixels are blurred (spread out across the exposure
surface) the motion vectors don't match up perfectly anymore. If you do
it in reverse (MB and then DOF) you have the same problem when the MB
pulls pixels across the exposure surface that wont match up to the
zdepth pass.
I suppose it might be worth looking into a unified solver for this that
does both at once. But I know of no such solution currently. Anyone
seen one?
Its probably worth mentioning that doing 2d DOF and MB both have another
issue in that they don't work with semi transparent objects all that
well either. Its needs to be fixed in the composite. And it usually
requires multiple depth and motion vector passes.
-brad
Stuart Hall wrote:
>I did my own little test to try it out the theory.
>
>There certainly is a difference. But as soon as motion blur comes into
>the equation I wonder whether either way makes that much of a
>difference.
>
>Does anyone have any opinions on motion blurring a depth pass?
>
>Cheers
>
>Stu
>
>-- --Original Message-- --
>From: owner-xsi@(protected) [mailto:owner-xsi@(protected)] On Behalf
>Of Raffaele "ThE_JacO" Fragapane
>Sent: 22 March 2005 11:07
>To: XSI@(protected)
>Subject: RE: DOF and Depth Pass problem
>
>For the record:
>Often (especially with hairy creatures moving along the camera rays) one
>channel of RGB proved to be not enough for us.
>
>In that case you can use a ray distance to drive a Gradient going
>through
>RGB and output 16bitpc, that way you get 65k*3 shades if you properly
>split
>the colour channels in compositing later on.
>
>You could also actually use 8bitpc output and create the whole 16 mills
>palette, but the rendertree we used got complicated and prone to little
>sampling glitches here and there.
>
>Btw a depth pass is better off non antialiased, because the depth pass
>will
>be AAed and sampled differently from the beauty pass.
>
>Rendering at doubled res without AA and resizing it down is a better
>solution, or eventually render both the beauty and the depth pass with
>non
>adaptive AA (min and max levels are the same), it could still present
>some
>artifacts but not a fraction of what you'd get from AAAed Beauty + AAAed
>Depth.
>
>~Raffaele Fragapane
>~Lead "I'm sure we can make it work"
>~Peerless Camera Company
>
>-- --Original Message-- --
>From: owner-xsi@(protected) [mailto:owner-xsi@(protected)] On Behalf
>Of
>Stuart Hall
>Sent: Tuesday, March 22, 2005 11:59 AM
>To: XSI@(protected)
>Subject: RE: DOF and Depth Pass problem
>
>On a completely different tack, I prefer to render out a separate depth
>pass as a 16-bit SGI because it is anti-aliased and can have motion
>blur. It doesn't take long and if the animation doesn't change it only
>has to be done the once.
>
>Just a thought.
>
>Stu
>
>
>---
>Outgoing mail is certified Virus Free.
>Checked by AVG anti-virus system (http://www.grisoft.com).
>Version: 6.0.799 / Virus Database: 543 - Release Date: 19-Nov-04 (See http://Nov-04.ora-code.com)
>
>
>---
>Unsubscribe? Mail Majordomo@(protected) with the following text in
>body:
>unsubscribe xsi
>
>---
>Unsubscribe? Mail Majordomo@(protected) with the following text in body:
>unsubscribe xsi
>
>
---
Unsubscribe? Mail Majordomo@(protected) with the following text in body:
unsubscribe xsi
|
|
|